linux/drivers/md/raid5.h
<<
>>
Prefs
   1#ifndef _RAID5_H
   2#define _RAID5_H
   3
   4#include <linux/raid/xor.h>
   5#include <linux/dmaengine.h>
   6
   7/*
   8 *
   9 * Each stripe contains one buffer per device.  Each buffer can be in
  10 * one of a number of states stored in "flags".  Changes between
  11 * these states happen *almost* exclusively under the protection of the
  12 * STRIPE_ACTIVE flag.  Some very specific changes can happen in bi_end_io, and
  13 * these are not protected by STRIPE_ACTIVE.
  14 *
  15 * The flag bits that are used to represent these states are:
  16 *   R5_UPTODATE and R5_LOCKED
  17 *
  18 * State Empty == !UPTODATE, !LOCK
  19 *        We have no data, and there is no active request
  20 * State Want == !UPTODATE, LOCK
  21 *        A read request is being submitted for this block
  22 * State Dirty == UPTODATE, LOCK
  23 *        Some new data is in this buffer, and it is being written out
  24 * State Clean == UPTODATE, !LOCK
  25 *        We have valid data which is the same as on disc
  26 *
  27 * The possible state transitions are:
  28 *
  29 *  Empty -> Want   - on read or write to get old data for  parity calc
  30 *  Empty -> Dirty  - on compute_parity to satisfy write/sync request.
  31 *  Empty -> Clean  - on compute_block when computing a block for failed drive
  32 *  Want  -> Empty  - on failed read
  33 *  Want  -> Clean  - on successful completion of read request
  34 *  Dirty -> Clean  - on successful completion of write request
  35 *  Dirty -> Clean  - on failed write
  36 *  Clean -> Dirty  - on compute_parity to satisfy write/sync (RECONSTRUCT or RMW)
  37 *
  38 * The Want->Empty, Want->Clean, Dirty->Clean, transitions
  39 * all happen in b_end_io at interrupt time.
  40 * Each sets the Uptodate bit before releasing the Lock bit.
  41 * This leaves one multi-stage transition:
  42 *    Want->Dirty->Clean
  43 * This is safe because thinking that a Clean buffer is actually dirty
  44 * will at worst delay some action, and the stripe will be scheduled
  45 * for attention after the transition is complete.
  46 *
  47 * There is one possibility that is not covered by these states.  That
  48 * is if one drive has failed and there is a spare being rebuilt.  We
  49 * can't distinguish between a clean block that has been generated
  50 * from parity calculations, and a clean block that has been
  51 * successfully written to the spare ( or to parity when resyncing).
  52 * To distingush these states we have a stripe bit STRIPE_INSYNC that
  53 * is set whenever a write is scheduled to the spare, or to the parity
  54 * disc if there is no spare.  A sync request clears this bit, and
  55 * when we find it set with no buffers locked, we know the sync is
  56 * complete.
  57 *
  58 * Buffers for the md device that arrive via make_request are attached
  59 * to the appropriate stripe in one of two lists linked on b_reqnext.
  60 * One list (bh_read) for read requests, one (bh_write) for write.
  61 * There should never be more than one buffer on the two lists
  62 * together, but we are not guaranteed of that so we allow for more.
  63 *
  64 * If a buffer is on the read list when the associated cache buffer is
  65 * Uptodate, the data is copied into the read buffer and it's b_end_io
  66 * routine is called.  This may happen in the end_request routine only
  67 * if the buffer has just successfully been read.  end_request should
  68 * remove the buffers from the list and then set the Uptodate bit on
  69 * the buffer.  Other threads may do this only if they first check
  70 * that the Uptodate bit is set.  Once they have checked that they may
  71 * take buffers off the read queue.
  72 *
  73 * When a buffer on the write list is committed for write it is copied
  74 * into the cache buffer, which is then marked dirty, and moved onto a
  75 * third list, the written list (bh_written).  Once both the parity
  76 * block and the cached buffer are successfully written, any buffer on
  77 * a written list can be returned with b_end_io.
  78 *
  79 * The write list and read list both act as fifos.  The read list,
  80 * write list and written list are protected by the device_lock.
  81 * The device_lock is only for list manipulations and will only be
  82 * held for a very short time.  It can be claimed from interrupts.
  83 *
  84 *
  85 * Stripes in the stripe cache can be on one of two lists (or on
  86 * neither).  The "inactive_list" contains stripes which are not
  87 * currently being used for any request.  They can freely be reused
  88 * for another stripe.  The "handle_list" contains stripes that need
  89 * to be handled in some way.  Both of these are fifo queues.  Each
  90 * stripe is also (potentially) linked to a hash bucket in the hash
  91 * table so that it can be found by sector number.  Stripes that are
  92 * not hashed must be on the inactive_list, and will normally be at
  93 * the front.  All stripes start life this way.
  94 *
  95 * The inactive_list, handle_list and hash bucket lists are all protected by the
  96 * device_lock.
  97 *  - stripes have a reference counter. If count==0, they are on a list.
  98 *  - If a stripe might need handling, STRIPE_HANDLE is set.
  99 *  - When refcount reaches zero, then if STRIPE_HANDLE it is put on
 100 *    handle_list else inactive_list
 101 *
 102 * This, combined with the fact that STRIPE_HANDLE is only ever
 103 * cleared while a stripe has a non-zero count means that if the
 104 * refcount is 0 and STRIPE_HANDLE is set, then it is on the
 105 * handle_list and if recount is 0 and STRIPE_HANDLE is not set, then
 106 * the stripe is on inactive_list.
 107 *
 108 * The possible transitions are:
 109 *  activate an unhashed/inactive stripe (get_active_stripe())
 110 *     lockdev check-hash unlink-stripe cnt++ clean-stripe hash-stripe unlockdev
 111 *  activate a hashed, possibly active stripe (get_active_stripe())
 112 *     lockdev check-hash if(!cnt++)unlink-stripe unlockdev
 113 *  attach a request to an active stripe (add_stripe_bh())
 114 *     lockdev attach-buffer unlockdev
 115 *  handle a stripe (handle_stripe())
 116 *     setSTRIPE_ACTIVE,  clrSTRIPE_HANDLE ...
 117 *              (lockdev check-buffers unlockdev) ..
 118 *              change-state ..
 119 *              record io/ops needed clearSTRIPE_ACTIVE schedule io/ops
 120 *  release an active stripe (release_stripe())
 121 *     lockdev if (!--cnt) { if  STRIPE_HANDLE, add to handle_list else add to inactive-list } unlockdev
 122 *
 123 * The refcount counts each thread that have activated the stripe,
 124 * plus raid5d if it is handling it, plus one for each active request
 125 * on a cached buffer, and plus one if the stripe is undergoing stripe
 126 * operations.
 127 *
 128 * The stripe operations are:
 129 * -copying data between the stripe cache and user application buffers
 130 * -computing blocks to save a disk access, or to recover a missing block
 131 * -updating the parity on a write operation (reconstruct write and
 132 *  read-modify-write)
 133 * -checking parity correctness
 134 * -running i/o to disk
 135 * These operations are carried out by raid5_run_ops which uses the async_tx
 136 * api to (optionally) offload operations to dedicated hardware engines.
 137 * When requesting an operation handle_stripe sets the pending bit for the
 138 * operation and increments the count.  raid5_run_ops is then run whenever
 139 * the count is non-zero.
 140 * There are some critical dependencies between the operations that prevent some
 141 * from being requested while another is in flight.
 142 * 1/ Parity check operations destroy the in cache version of the parity block,
 143 *    so we prevent parity dependent operations like writes and compute_blocks
 144 *    from starting while a check is in progress.  Some dma engines can perform
 145 *    the check without damaging the parity block, in these cases the parity
 146 *    block is re-marked up to date (assuming the check was successful) and is
 147 *    not re-read from disk.
 148 * 2/ When a write operation is requested we immediately lock the affected
 149 *    blocks, and mark them as not up to date.  This causes new read requests
 150 *    to be held off, as well as parity checks and compute block operations.
 151 * 3/ Once a compute block operation has been requested handle_stripe treats
 152 *    that block as if it is up to date.  raid5_run_ops guaruntees that any
 153 *    operation that is dependent on the compute block result is initiated after
 154 *    the compute block completes.
 155 */
 156
 157/*
 158 * Operations state - intermediate states that are visible outside of 
 159 *   STRIPE_ACTIVE.
 160 * In general _idle indicates nothing is running, _run indicates a data
 161 * processing operation is active, and _result means the data processing result
 162 * is stable and can be acted upon.  For simple operations like biofill and
 163 * compute that only have an _idle and _run state they are indicated with
 164 * sh->state flags (STRIPE_BIOFILL_RUN and STRIPE_COMPUTE_RUN)
 165 */
 166/**
 167 * enum check_states - handles syncing / repairing a stripe
 168 * @check_state_idle - check operations are quiesced
 169 * @check_state_run - check operation is running
 170 * @check_state_result - set outside lock when check result is valid
 171 * @check_state_compute_run - check failed and we are repairing
 172 * @check_state_compute_result - set outside lock when compute result is valid
 173 */
 174enum check_states {
 175        check_state_idle = 0,
 176        check_state_run, /* xor parity check */
 177        check_state_run_q, /* q-parity check */
 178        check_state_run_pq, /* pq dual parity check */
 179        check_state_check_result,
 180        check_state_compute_run, /* parity repair */
 181        check_state_compute_result,
 182};
 183
 184/**
 185 * enum reconstruct_states - handles writing or expanding a stripe
 186 */
 187enum reconstruct_states {
 188        reconstruct_state_idle = 0,
 189        reconstruct_state_prexor_drain_run,     /* prexor-write */
 190        reconstruct_state_drain_run,            /* write */
 191        reconstruct_state_run,                  /* expand */
 192        reconstruct_state_prexor_drain_result,
 193        reconstruct_state_drain_result,
 194        reconstruct_state_result,
 195};
 196
 197struct stripe_head {
 198        struct hlist_node       hash;
 199        struct list_head        lru;          /* inactive_list or handle_list */
 200        struct r5conf           *raid_conf;
 201        short                   generation;     /* increments with every
 202                                                 * reshape */
 203        sector_t                sector;         /* sector of this row */
 204        short                   pd_idx;         /* parity disk index */
 205        short                   qd_idx;         /* 'Q' disk index for raid6 */
 206        short                   ddf_layout;/* use DDF ordering to calculate Q */
 207        unsigned long           state;          /* state flags */
 208        atomic_t                count;        /* nr of active thread/requests */
 209        int                     bm_seq; /* sequence number for bitmap flushes */
 210        int                     disks;          /* disks in stripe */
 211        enum check_states       check_state;
 212        enum reconstruct_states reconstruct_state;
 213        spinlock_t              stripe_lock;
 214        /**
 215         * struct stripe_operations
 216         * @target - STRIPE_OP_COMPUTE_BLK target
 217         * @target2 - 2nd compute target in the raid6 case
 218         * @zero_sum_result - P and Q verification flags
 219         * @request - async service request flags for raid_run_ops
 220         */
 221        struct stripe_operations {
 222                int                  target, target2;
 223                enum sum_check_flags zero_sum_result;
 224                #ifdef CONFIG_MULTICORE_RAID456
 225                unsigned long        request;
 226                wait_queue_head_t    wait_for_ops;
 227                #endif
 228        } ops;
 229        struct r5dev {
 230                /* rreq and rvec are used for the replacement device when
 231                 * writing data to both devices.
 232                 */
 233                struct bio      req, rreq;
 234                struct bio_vec  vec, rvec;
 235                struct page     *page;
 236                struct bio      *toread, *read, *towrite, *written;
 237                sector_t        sector;                 /* sector of this page */
 238                unsigned long   flags;
 239        } dev[1]; /* allocated with extra space depending of RAID geometry */
 240};
 241
 242/* stripe_head_state - collects and tracks the dynamic state of a stripe_head
 243 *     for handle_stripe.
 244 */
 245struct stripe_head_state {
 246        /* 'syncing' means that we need to read all devices, either
 247         * to check/correct parity, or to reconstruct a missing device.
 248         * 'replacing' means we are replacing one or more drives and
 249         * the source is valid at this point so we don't need to
 250         * read all devices, just the replacement targets.
 251         */
 252        int syncing, expanding, expanded, replacing;
 253        int locked, uptodate, to_read, to_write, failed, written;
 254        int to_fill, compute, req_compute, non_overwrite;
 255        int failed_num[2];
 256        int p_failed, q_failed;
 257        int dec_preread_active;
 258        unsigned long ops_request;
 259
 260        struct bio *return_bi;
 261        struct md_rdev *blocked_rdev;
 262        int handle_bad_blocks;
 263};
 264
 265/* Flags for struct r5dev.flags */
 266enum r5dev_flags {
 267        R5_UPTODATE,    /* page contains current data */
 268        R5_LOCKED,      /* IO has been submitted on "req" */
 269        R5_DOUBLE_LOCKED,/* Cannot clear R5_LOCKED until 2 writes complete */
 270        R5_OVERWRITE,   /* towrite covers whole page */
 271/* and some that are internal to handle_stripe */
 272        R5_Insync,      /* rdev && rdev->in_sync at start */
 273        R5_Wantread,    /* want to schedule a read */
 274        R5_Wantwrite,
 275        R5_Overlap,     /* There is a pending overlapping request
 276                         * on this block */
 277        R5_ReadNoMerge, /* prevent bio from merging in block-layer */
 278        R5_ReadError,   /* seen a read error here recently */
 279        R5_ReWrite,     /* have tried to over-write the readerror */
 280
 281        R5_Expanded,    /* This block now has post-expand data */
 282        R5_Wantcompute, /* compute_block in progress treat as
 283                         * uptodate
 284                         */
 285        R5_Wantfill,    /* dev->toread contains a bio that needs
 286                         * filling
 287                         */
 288        R5_Wantdrain,   /* dev->towrite needs to be drained */
 289        R5_WantFUA,     /* Write should be FUA */
 290        R5_SyncIO,      /* The IO is sync */
 291        R5_WriteError,  /* got a write error - need to record it */
 292        R5_MadeGood,    /* A bad block has been fixed by writing to it */
 293        R5_ReadRepl,    /* Will/did read from replacement rather than orig */
 294        R5_MadeGoodRepl,/* A bad block on the replacement device has been
 295                         * fixed by writing to it */
 296        R5_NeedReplace, /* This device has a replacement which is not
 297                         * up-to-date at this stripe. */
 298        R5_WantReplace, /* We need to update the replacement, we have read
 299                         * data in, and now is a good time to write it out.
 300                         */
 301};
 302
 303/*
 304 * Stripe state
 305 */
 306enum {
 307        STRIPE_ACTIVE,
 308        STRIPE_HANDLE,
 309        STRIPE_SYNC_REQUESTED,
 310        STRIPE_SYNCING,
 311        STRIPE_INSYNC,
 312        STRIPE_PREREAD_ACTIVE,
 313        STRIPE_DELAYED,
 314        STRIPE_DEGRADED,
 315        STRIPE_BIT_DELAY,
 316        STRIPE_EXPANDING,
 317        STRIPE_EXPAND_SOURCE,
 318        STRIPE_EXPAND_READY,
 319        STRIPE_IO_STARTED,      /* do not count towards 'bypass_count' */
 320        STRIPE_FULL_WRITE,      /* all blocks are set to be overwritten */
 321        STRIPE_BIOFILL_RUN,
 322        STRIPE_COMPUTE_RUN,
 323        STRIPE_OPS_REQ_PENDING,
 324        STRIPE_ON_UNPLUG_LIST,
 325};
 326
 327/*
 328 * Operation request flags
 329 */
 330enum {
 331        STRIPE_OP_BIOFILL,
 332        STRIPE_OP_COMPUTE_BLK,
 333        STRIPE_OP_PREXOR,
 334        STRIPE_OP_BIODRAIN,
 335        STRIPE_OP_RECONSTRUCT,
 336        STRIPE_OP_CHECK,
 337};
 338/*
 339 * Plugging:
 340 *
 341 * To improve write throughput, we need to delay the handling of some
 342 * stripes until there has been a chance that several write requests
 343 * for the one stripe have all been collected.
 344 * In particular, any write request that would require pre-reading
 345 * is put on a "delayed" queue until there are no stripes currently
 346 * in a pre-read phase.  Further, if the "delayed" queue is empty when
 347 * a stripe is put on it then we "plug" the queue and do not process it
 348 * until an unplug call is made. (the unplug_io_fn() is called).
 349 *
 350 * When preread is initiated on a stripe, we set PREREAD_ACTIVE and add
 351 * it to the count of prereading stripes.
 352 * When write is initiated, or the stripe refcnt == 0 (just in case) we
 353 * clear the PREREAD_ACTIVE flag and decrement the count
 354 * Whenever the 'handle' queue is empty and the device is not plugged, we
 355 * move any strips from delayed to handle and clear the DELAYED flag and set
 356 * PREREAD_ACTIVE.
 357 * In stripe_handle, if we find pre-reading is necessary, we do it if
 358 * PREREAD_ACTIVE is set, else we set DELAYED which will send it to the delayed queue.
 359 * HANDLE gets cleared if stripe_handle leaves nothing locked.
 360 */
 361
 362
 363struct disk_info {
 364        struct md_rdev  *rdev, *replacement;
 365};
 366
 367struct r5conf {
 368        struct hlist_head       *stripe_hashtbl;
 369        struct mddev            *mddev;
 370        int                     chunk_sectors;
 371        int                     level, algorithm;
 372        int                     max_degraded;
 373        int                     raid_disks;
 374        int                     max_nr_stripes;
 375
 376        /* reshape_progress is the leading edge of a 'reshape'
 377         * It has value MaxSector when no reshape is happening
 378         * If delta_disks < 0, it is the last sector we started work on,
 379         * else is it the next sector to work on.
 380         */
 381        sector_t                reshape_progress;
 382        /* reshape_safe is the trailing edge of a reshape.  We know that
 383         * before (or after) this address, all reshape has completed.
 384         */
 385        sector_t                reshape_safe;
 386        int                     previous_raid_disks;
 387        int                     prev_chunk_sectors;
 388        int                     prev_algo;
 389        short                   generation; /* increments with every reshape */
 390        unsigned long           reshape_checkpoint; /* Time we last updated
 391                                                     * metadata */
 392        long long               min_offset_diff; /* minimum difference between
 393                                                  * data_offset and
 394                                                  * new_data_offset across all
 395                                                  * devices.  May be negative,
 396                                                  * but is closest to zero.
 397                                                  */
 398
 399        struct list_head        handle_list; /* stripes needing handling */
 400        struct list_head        hold_list; /* preread ready stripes */
 401        struct list_head        delayed_list; /* stripes that have plugged requests */
 402        struct list_head        bitmap_list; /* stripes delaying awaiting bitmap update */
 403        struct bio              *retry_read_aligned; /* currently retrying aligned bios   */
 404        struct bio              *retry_read_aligned_list; /* aligned bios retry list  */
 405        atomic_t                preread_active_stripes; /* stripes with scheduled io */
 406        atomic_t                active_aligned_reads;
 407        atomic_t                pending_full_writes; /* full write backlog */
 408        int                     bypass_count; /* bypassed prereads */
 409        int                     bypass_threshold; /* preread nice */
 410        struct list_head        *last_hold; /* detect hold_list promotions */
 411
 412        atomic_t                reshape_stripes; /* stripes with pending writes for reshape */
 413        /* unfortunately we need two cache names as we temporarily have
 414         * two caches.
 415         */
 416        int                     active_name;
 417        char                    cache_name[2][32];
 418        struct kmem_cache               *slab_cache; /* for allocating stripes */
 419
 420        int                     seq_flush, seq_write;
 421        int                     quiesce;
 422
 423        int                     fullsync;  /* set to 1 if a full sync is needed,
 424                                            * (fresh device added).
 425                                            * Cleared when a sync completes.
 426                                            */
 427        int                     recovery_disabled;
 428        /* per cpu variables */
 429        struct raid5_percpu {
 430                struct page     *spare_page; /* Used when checking P/Q in raid6 */
 431                void            *scribble;   /* space for constructing buffer
 432                                              * lists and performing address
 433                                              * conversions
 434                                              */
 435        } __percpu *percpu;
 436        size_t                  scribble_len; /* size of scribble region must be
 437                                               * associated with conf to handle
 438                                               * cpu hotplug while reshaping
 439                                               */
 440#ifdef CONFIG_HOTPLUG_CPU
 441        struct notifier_block   cpu_notify;
 442#endif
 443
 444        /*
 445         * Free stripes pool
 446         */
 447        atomic_t                active_stripes;
 448        struct list_head        inactive_list;
 449        wait_queue_head_t       wait_for_stripe;
 450        wait_queue_head_t       wait_for_overlap;
 451        int                     inactive_blocked;       /* release of inactive stripes blocked,
 452                                                         * waiting for 25% to be free
 453                                                         */
 454        int                     pool_size; /* number of disks in stripeheads in pool */
 455        spinlock_t              device_lock;
 456        struct disk_info        *disks;
 457
 458        /* When taking over an array from a different personality, we store
 459         * the new thread here until we fully activate the array.
 460         */
 461        struct md_thread        *thread;
 462};
 463
 464/*
 465 * Our supported algorithms
 466 */
 467#define ALGORITHM_LEFT_ASYMMETRIC       0 /* Rotating Parity N with Data Restart */
 468#define ALGORITHM_RIGHT_ASYMMETRIC      1 /* Rotating Parity 0 with Data Restart */
 469#define ALGORITHM_LEFT_SYMMETRIC        2 /* Rotating Parity N with Data Continuation */
 470#define ALGORITHM_RIGHT_SYMMETRIC       3 /* Rotating Parity 0 with Data Continuation */
 471
 472/* Define non-rotating (raid4) algorithms.  These allow
 473 * conversion of raid4 to raid5.
 474 */
 475#define ALGORITHM_PARITY_0              4 /* P or P,Q are initial devices */
 476#define ALGORITHM_PARITY_N              5 /* P or P,Q are final devices. */
 477
 478/* DDF RAID6 layouts differ from md/raid6 layouts in two ways.
 479 * Firstly, the exact positioning of the parity block is slightly
 480 * different between the 'LEFT_*' modes of md and the "_N_*" modes
 481 * of DDF.
 482 * Secondly, or order of datablocks over which the Q syndrome is computed
 483 * is different.
 484 * Consequently we have different layouts for DDF/raid6 than md/raid6.
 485 * These layouts are from the DDFv1.2 spec.
 486 * Interestingly DDFv1.2-Errata-A does not specify N_CONTINUE but
 487 * leaves RLQ=3 as 'Vendor Specific'
 488 */
 489
 490#define ALGORITHM_ROTATING_ZERO_RESTART 8 /* DDF PRL=6 RLQ=1 */
 491#define ALGORITHM_ROTATING_N_RESTART    9 /* DDF PRL=6 RLQ=2 */
 492#define ALGORITHM_ROTATING_N_CONTINUE   10 /*DDF PRL=6 RLQ=3 */
 493
 494
 495/* For every RAID5 algorithm we define a RAID6 algorithm
 496 * with exactly the same layout for data and parity, and
 497 * with the Q block always on the last device (N-1).
 498 * This allows trivial conversion from RAID5 to RAID6
 499 */
 500#define ALGORITHM_LEFT_ASYMMETRIC_6     16
 501#define ALGORITHM_RIGHT_ASYMMETRIC_6    17
 502#define ALGORITHM_LEFT_SYMMETRIC_6      18
 503#define ALGORITHM_RIGHT_SYMMETRIC_6     19
 504#define ALGORITHM_PARITY_0_6            20
 505#define ALGORITHM_PARITY_N_6            ALGORITHM_PARITY_N
 506
 507static inline int algorithm_valid_raid5(int layout)
 508{
 509        return (layout >= 0) &&
 510                (layout <= 5);
 511}
 512static inline int algorithm_valid_raid6(int layout)
 513{
 514        return (layout >= 0 && layout <= 5)
 515                ||
 516                (layout >= 8 && layout <= 10)
 517                ||
 518                (layout >= 16 && layout <= 20);
 519}
 520
 521static inline int algorithm_is_DDF(int layout)
 522{
 523        return layout >= 8 && layout <= 10;
 524}
 525
 526extern int md_raid5_congested(struct mddev *mddev, int bits);
 527extern void md_raid5_kick_device(struct r5conf *conf);
 528extern int raid5_set_cache_size(struct mddev *mddev, int size);
 529#endif
 530
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.